Adipose Tissue Inflammation, Oxidative Stress, and Altered Adipogenesis Are Associated With Dyslipidemia in Obesity: A Multiomics Profiling Study

Abstract

Background

Obesity is an important risk factor for cardiometabolic disease, including dyslipidemia and atherosclerotic cardiovascular disease. Although the role of the liver in dyslipidemia is established, the contribution of adipose tissue is less clear. This study aims to clarify the role of adipose tissue in lipid metabolism and dyslipidemia.

Methods

We conducted a cross‐sectional analysis of 125 patients from the BARIA (The Immune System and Microbial Tone in Relation to NAFLD/NASH Before and After Bariatric Surgery in the Morbidly Obese in Amsterdam) longitudinal cohort study undergoing bariatric surgery. Comprehensive phenotyping included fasting untargeted plasma metabolomics, lipid, lipoprotein, adipokine profiling, RNA sequencing, and fecal shotgun metagenomics. Tissue transcriptomic and plasma metabolites were compared between individuals with and without dyslipidemia.

Results

Dyslipidemia was present in 43 of 125 individuals (34.4%), with higher triglycerides (1.62 versus 1.24 mmol/L), apoB (apolipoprotein B; 93.15 versus 81.81 mg/dL), and lower high‐density lipoprotein (1.02 versus 1.35 mmol/L) and apoAI (136.40 versus 161.35 mg/dL). Plasma adipokines showed limited differences: leptin concentrations were lower in dyslipidemia in unadjusted analysis but reduced after adjustment for age, sex, and body weight (adjusted P=0.057). RNA sequencing identified altered gene expression of liver, jejunum, visceral and subcutaneous adipose tissue, most pronounced in subcutaneous adipose tissue. Dyslipidemia was associated with adipose tissue pathways related to inflammation, oxidative stress, and adipogenesis. Plasma metabolomics revealed associations with endocannabinoid‐like, secondary bile acid, plasmalogen, butyrate, and sphingolipid metabolites. Gut metagenome analysis found modest differences.

Conclusions

Dyslipidemia in obesity is associated with transcriptomic alterations in adipose tissue, including subcutaneous adipose tissue, involving inflammation, oxidative stress, and adipogenesis. These findings support a role of adipose tissue in lipid regulation beyond hepatic pathways.

Nonstandard Abbreviations and Acronyms

AT

adipose tissue

GM

gut microbiota

KEGG

Kyoto Encyclopedia of Genes and Genomes

ROS

reactive oxygen species

SAT

subcutaneous adipose tissue

SCFA

short‐chain fatty acid

VAT

visceral adipose tissue

Research Perspective.

What Is New ?

• This study demonstrates that dyslipidemia in obesity is associated with distinct adipose tissue molecular signatures, including coordinated alterations in inflammation, oxidative stress, and adipogenesis across both visceral and subcutaneous fat depots. These findings highlight adipose tissue, including subcutaneous fat, as a potential contributor to dyslipidemia in obesity and suggest that lipid abnormalities may reflect adipose tissue dysfunction rather than hepatic lipid metabolism alone.

What Questions Should be Addressed Next?

• Future studies should determine whether adipose tissue‐specific molecular signatures can identify individuals with obesity at higher cardiometabolic risk and whether mechanistic and translational approaches targeting adipose tissue inflammation or remodeling can improve lipid profiles and cardiovascular outcomes.

Obesity is increasing worldwide and for individuals with a high body mass index (BMI), cardiovascular disease remains the most attributed cause of death. ¹ , ² Adipose tissue (AT) is not only an energy storing organ but also a metabolic and endocrinologically active organ and a driver of cardiometabolic diseases like insulin resistance and type 2 diabetes, hypertension, metabolic dysfunction‐associated steatotic liver disease and atherosclerotic cardiovascular disease. ³ , ⁴ , ⁵ Obesity not only contributes to insulin resistance, but also directly to dyslipidemia, marked by decreased high‐density lipoprotein cholesterol (HDL‐C) and increased plasma triglyceride concentration through enhanced hepatic release of apoB (apolipoprotein B)‐containing lipoproteins, such as very low‐density lipoprotein (VLDL). ⁶ The mechanism behind the development of dyslipidemia in obesity is only partly understood, and studies have shown that the degree of obesity and insulin resistance does not have a linear relationship with dyslipidemia; individuals with overweight (BMI 25.0–29.9 kg/m²) may have more significant changes in their lipid profile as compared to individuals with class III obesity (BMI >40 kg/m²). ⁷ , ⁸ This indicates that besides pathogenic mechanisms, also protective ones could be at play in obesity, influencing susceptibility to develop dyslipidemia and subsequent atherosclerotic cardiovascular disease.

Although the role of the liver in lipid metabolism is undisputed, AT is increasingly seen as another important player. ⁹ According to population‐based and cohort studies, the amount of visceral adipose tissue (VAT) appears to be a stronger factor for the development of dyslipidemia and associated cardiovascular disease risk than subcutaneous adipose tissue (SAT). ¹⁰ , ¹¹ , ¹² Apart from the volume of AT, the function of AT, or rather, AT dysfunction, appears to influence glucose and lipid metabolism most. ⁴ AT dysfunction is marked not only by adipocyte hypertrophy but also by AT inflammation (M1‐macrophage predominance), oxidative stress, and alterations in the adipokine profile such as increased leptin and decreased adiponectin concentrations. ⁶ , ¹³ , ¹⁴ , ¹⁵ Changes in adipokine release is relevant, because, on the one hand, adipokines such as adiponectin, omentin, and FGF‐21 (fibroblast growth factor 21) have insulin sensitizing, anti‐inflammatory, and anti‐atherogenic properties, whereas leptin, resistin, and visfatin are thought to be detrimental for health. ⁹ , ¹⁴ , ¹⁶ One of the suggested mechanisms contributing to a dysfunctional AT is hypoxia and oxidative stress caused by AT hypertrophy exceeding its vascularization potential. ¹⁷

As mentioned, not all individuals with excess AT develop dyslipidemia, hence further characterization of the role of AT in lipid metabolism is necessary. By comparing AT and liver gene expression patterns and metabolites of individuals with obesity and dyslipidemia versus individuals with obesity but without dyslipidemia, both protective and pathogenic mechanisms by which AT interacts with other tissues can be uncovered. Gut microbiota and its produced metabolites are also potential players in obesity, dyslipidemia, and atherosclerotic cardiovascular disease. ⁶ , ¹⁸

Therefore, in this cross‐sectional study we hypothesized that besides liver, different AT depots (specifically VAT) also have significant up‐ or downregulated genetic pathways in association with dyslipidemia. We thus postulate that AT in individuals with obesity can undergo 1 of 2 distinct pathways: 1 involving excess of AT (hyperplasia) but with maintenance of lipid homeostasis, and the other involving AT dysfunction and the development of dyslipidemia. We studied this hypothesis in a cohort of patients with obesity scheduled for bariatric surgery in which comprehensive metabolic phenotyping was performed, including measurements of insulin resistance, tissue RNA, plasma metabolites, lipids, and adipokines, as well as determination of the gut microbiota (GM) composition. We aimed to gain insight into pathways and genes involved in dyslipidemia in individuals with obesity, specifically those of VAT and SAT, and we specifically focused on genes and adipokines proposed to be involved in AT dysfunction such as inflammation and oxidative stress.

METHODS

Study Design and Population

We used blood, fecal, and tissue biopsies of jejunum, liver, as well as AT (subcutaneous and visceral) obtained during surgery from selected patients of the BARIA (The Immune System and Microbial Tone in Relation to NAFLD/NASH Before and After Bariatric Surgery in bese in Amsterdam) longitudinal cohort study. ¹⁹ The BARIA cohort consists of individuals with obesity undergoing bariatric surgery with a BMI of ≥40 kg/m² or ≥35 kg/m² with obesity‐related comorbidities including hypertension, type 2 diabetes, obstructive sleep apnea, osteoarthritis, or dyslipidemia (primary forms excluded). Patients for this study were recruited into the BARIA cohort between 2016 and 2021 at the former Slotervaart Ziekenhuis (Amsterdam, the Netherlands) and Spaarne Gasthuis (Hoofddorp, the Netherlands). Samples were further analyzed through tissue transcriptomics, untargeted plasma metabolomics, and fecal microbial DNA sequencing.

Due to potential confounding effects of medication use, patients on lipid‐lowering or antidiabetic agents were excluded from analysis. This resulted in 125 BARIA cohort patients who had both a blood sample drawn on the morning of their bariatric surgery and undergone biopsy of all aforementioned tissues (Figure S1). Among these 125, we divided patients by the presence of dyslipidemia or not, where dyslipidemia was defined by having ≥1 of the National Health and Nutrition Examination Survey Adult Treatment Panel III criteria (criteria are described in the Supporting Information). ²⁰ Besides dichotomous comparisons (dyslipidemia versus no dyslipidemia), we also performed association analyses using continuous lipid or lipoprotein concentrations to capture graded relationships that may not be found when using a binary definition alone.

The research was conducted in accordance with the Declaration of Helsinki and had approval from the Ethics Committee of the Amsterdam University Medical Center (Trialregister NL5837). All study participants gave written informed consent before participation.

Data Availability

Data are available upon reasonable request through contacting the corresponding author.

Sample Collection

Baseline characteristics were obtained after an overnight fast on the participant's preoperative visit. Participants collected stool samples on the day before their planned bariatric surgery. Upon arrival at the admission department for surgery, fecal samples were instantly frozen at −80 °C. For plasma lipids, lipoproteins, and adipokines, we obtained venous blood samples on EDTA‐coated tubes on the morning of bariatric surgery, with centrifugation at 4 °C for 15 minutes, we isolated the plasma fraction and stored it at −80 °C until further processing. Biopsies of SAT, VAT, liver, and jejunum were performed perioperatively and snap frozen in liquid nitrogen within 20 seconds of biopsy, ensuring minimal RNA breakdown. Biopsies were obtained within 2 hours of blood drawing.

Plasma Sample Analysis

For measurement of total cholesterol, triglycerides, HDL‐C, apo‐I, apoB, we used commercially available kits (DiaSys, WAKO) on an automated clinical chemistry analyzer (Selectra, Sopachem, the Netherlands). We determined LDL‐C with the Friedewald equation. ²¹ We further investigated the concept of adipocyte dysfunction through measurement of adipokines and determined the plasma adiponectin‐leptin ratio as a marker of dyslipidemia. Plasma adiponectin and leptin concentrations were measured using human Adiponectin/Acrp30 DuoSet ELISA (Y1065‐05; R&D Systems, Minneapolis, MN, USA) and human Leptin DuoSet ELISA (DY398‐05; R&D Systems, Minneapolis, MN, USA).

RNA Sequencing of Tissue Samples

Tissue transcriptomics methods are explained in detail in Supporting Information Methods. Sample sizes for tissue transcriptomics were SAT n=116 (Figure S2), visceral fat n=124, liver n=123, and jejunum n=118. Analyses were adjusted for potential confounders: age, sex, and sequencing batch. Next, we used the pathfindR R package version 2.3.0 ²² for pathway enrichment analysis, and Kyoto Encyclopedia of Genes and Genomes (KEGG) database for biological pathway identification.

Plasma Metabolomics

Untargeted plasma metabolomics were done at Metabolon (USA). We initially applied a binary approach (dyslipidemia present or not) and sought correlations with certain metabolites on the presence of dyslipidemia. Next, we performed correlation analyses using continuous values of apoAI, apoB, triglycerides, cholesterol, HDL‐C, and LDL‐C in day‐of‐surgery blood samples. This 2‐step approach was used to identify metabolites associated with dyslipidemia status and to capture metabolites that vary in relation to lipid burden across individuals. Full details of metabolomics methodology are provided in the Supporting Information Methods.

Gut Bacterial Metagenomics

For gut bacterial metagenomics we applied a binary approach (dyslipidemia being present or not) only in relation to GM composition. DNA microbial metagenome sequencing data of 110 fecal samples were analyzed. Methods are detailed in the Supporting Information Methods.

Multiomics Integration Analysis

To integrate tissue transcriptomics with plasma metabolomics, we applied correlation‐based network analysis between dyslipidemia‐associated genes in SAT, VAT, liver, and jejunum with dyslipidemia‐associated metabolites. Significant associations were visualized as networks.

Statistical Analysis

Statistical analyses were performed using R v4.3.03 ²³ and RStudio 2024.04.2. ²⁴ Normally distributed data were described with mean±SD, nonnormal with median (first–third quartile). Categorical variables were compared using the chi‐square test or Fisher exact test, as appropriate. We compared normally distributed continuous variables with Student t test, and nonnormal with Mann–Whitney U testing. Correlations of normally distributed continuous data were done with Pearson's correlation and nonnormal data with Spearman's correlation. Because we correlated multiple adipokines and plasma lipid, and lipoprotein levels, to reduce the false discovery rate (FDR), we applied the Benjamini–Hochberg correction to results of transcriptomic analysis and plasma biomarkers, repeated correlations were filtered for significant associations using the same adjustment method. For analyses of circulating adipokines in relation to dyslipidemia status, we used multiple linear regression adjusting for prespecified covariates selected a priori on the basis of biological and clinical relevance. For transcriptomics, besides total population, subgroup analyses were also performed by sex using stratified differential expression models (male and female separately), with the same covariate adjustment as in the primary analyses. For transcriptomics analyses, statistical significance was defined using FDR‐adjusted P values. For biological interpretation in the main text, emphasis was placed on genes with larger effect sizes (log₂ fold change ≥0.5 for DESeq2 analyses of dyslipidemia status or absolute β‐coefficient ≥0.3 for microbiome multivariable association with linear models of continuous lipid associations), and associations with smaller effect sizes were retained in the Tables S1 and S2 for completeness. For all statistical comparisons and tests, we considered a statistical significance level of P<0.05 as significant.

RESULTS

A visual summary of the principal transcriptomic and metabolomic alterations associated with dyslipidemia and plasma lipid profiles is given in Figure 1.

Figure 1. Graphical abstract.

A, Study population. B, Experimental approach. C, Summary of transcriptomic and metabolomic results. Genes expressed in SAT, VAT, or liver with positive (+) or negative (−) associations with dyslipidemia or an adverse lipid profile (elevated triglycerides or decreased HDL‐C or apoAI). Genes described in detail in Tables S1 and S2. *These genes have an inverse relationship with HDL‐C, meaning if positive (+), they were positively related to dyslipidemia (negative correlation with HDL‐C), and if negative (−), they were negatively related to dyslipidemia (positive correlation with HDL‐C). Created in BioRender. Zwartjes, M. (2026) https://BioRender.com/nlgzhcl. ApoAI indicates apolipoprotein AI; AT, adipose tissue; HDL‐C, high‐density lipoprotein cholesterol; HIF, hypoxia‐inducible factor 1; isoUDCA, isoursodeoxycholate; MAFLD, metabolic associated fatty liver disease; PPARα, peroxisome proliferator‐activated receptor alpha; ROS, reactive oxygen species; SAT, subcutaneous adipose tissue; TNF‐α, tumor necrosis factor alpha; and VAT, viscous adipose tissue.

Patients and Plasma Biomarkers

We included 125 patients, with a mean BMI of 39.5 kg/m², of whom 80.8% were female. Dyslipidemia was present in 43 patients (34.4%), whereas 82 (65.6%) had a normal lipid profile. Although body weight was modestly higher in the dyslipidemia group, adiposity was comparable between groups, as BMI did not differ between patients with and without dyslipidemia. The study population was ethnically homogeneous, and ethnicity distribution did not differ between groups (P=0.17). Patient characteristics are shown in Table 1.

Table 1.

Patient Characteristics

	Total population (n=125, 100%)	Dyslipidemia (n=43, 34.4%)	No dyslipidemia (n=82, 65.6%)	P value
Demographic
Female sex	101 (80.8%)	28 (65.1%)	73 (89.0%)	<0.001‡
Age, y	44.3 ± 10.4	42.6 ± 10.6	42.3 ± 10.3	0.176
Race
White	108 (86.4%)	40 (93.0%)	68 (82.9%)	0.170*
Other	17 (13.6%)	3 (7.0%)	14 (17.1%)
Anthropometric
Weight, kg	115.0 ± 13.8	122.2 ± 14.4	112.4 ± 13.8	0.004‡
Body mass index, kg/m2	39.5 ± 3.6	39.6 ± 3.1	39.4 ± 3.8	0.800
Waist circumference, cm	121.4 ± 10.1	123.7 ± 10.8	120.1 ± 9.5	0.058
Hip circumference, cm	130.7 ± 9.4	130.6 ± 9.4	130.8 ± 9.5	0.905
Waist‐to‐hip ratio	0.9 ± 0.1	0.9 ± 0.1	0.9 ± 0.1	0.056
Body fat, kg	53.7 ± 8.4	54.6 ± 7.5	53.29 ± 8.9	0.470
Systolic blood pressure, mm Hg	122.0 ± 15.0	122.0 ± 15.0	121.0 ± 15.1	0.853
Diastolic blood pressure, mm Hg	79.1 ± 9.4	77.1 ± 8.6	80.1 ± 9.6	0.093
Heart rate, bpm	73.9 ± 11.0	70.6 ± 9.9	75.7 ± 11.2	0.011‡
Clinical laboratory values at baseline
Alkaline phosphatase, U/L	83.8 ± 21.7	84.4 ± 21.0	83.5 ± 22.2	0.826
γ‐glutamyltransferase, U/L	28.0 (20.0–41.0)	27.0 (19.5–41.0)	28.0 (20.0–39.75)	0.920
Aspartate transaminase, U/L	23.0 (20.0–28.0)	25.0 (21.0–29.0)	23.0 (20.0–27.5)	0.144
Alanine transaminase, U/L	30.0 (22.0–38.5)	29.0 (22.0–37.0)	31.0 (22.0–39.3)	0.609
Ferritin μg/L	85 (47.5–171.5)	131 (81.0–196.0)	75.5 (33.8–134.8)	0.002‡
C=reactive protein, mg/L	6.8 (3.5–9.0)	5.0 (3.0–9.0)	7.4 (4.0–9.0)	0.160
Hemoglobin A1C baseline %	5.41 ± 0.4	5.41 ± 0.3	5.42 ± 0.4	0.951
Baseline mixed meal test
Fasting glucose mmol/L	5.67 ± 0.6	5.72 ± 0.5	5.64 ± 0.7	0.458
Fasting insulin mIU/L	82.65 ± 47.1	84.87 ± 36.3	81.52 ± 52.0	0.709
C‐peptide ng/mL	0.76 ± 0.3	0.82 ± 0.3	0.73 ± 0.3	0.124
HOMA‐1 IR	2.79 (1.80–3.98)	3.09 (2.29–4.27)	2.37 (1.63–3.86)	0.123
HOMA‐2 IR	1.49 (0.98–2.09)	1.69 (1.35–2.16)	1.29 (0.97–2.02)	0.068
Liver histology
No steatosis	58 (47.2%)	17 (39.5%)	41 (51.2%)	0.434†
Metabolic dysfunction‐associated steatotic liver disease	49 (39.8%)	19 (44.2%)	30 (37.5%)
Metabolic dysfunction‐associated steatohepatitis	16 (13.0%)	7 (16.3%)	9 (11.2%)

Data expressed as mean ± SD or as median (first–third quartile). For calculating HOMA, we used “HOMA‐1=(fasting plasma glucose [mmol/L] × fasting plasma insulin [μU/mL] x)/22.5”, ⁶⁹ and to acquire HOMA‐2 values, ⁷⁰ we used the online available HOMA‐2 calculator software. ⁷¹

HOMA IR indicates homeostatic model assessment for insulin resistance.

^* P value of Fisher exact test.

^† P value of χ ² test for independence.

^‡ Significant findings with P<0.05.

Complete lipid and lipoprotein concentrations are seen in Table 2. Patients with dyslipidemia had lower HDL‐C and apoAI concentrations and higher triglycerides, apoB concentrations, whereas LDL‐C did not differ in patients with and without dyslipidemia. With respect to adipokines, adiponectin concentrations did not differ between groups (P=0.072). Leptin concentrations were lower in patients with dyslipidemia in unadjusted analyses (P=0.010), whereas the adiponectin‐to‐leptin ratio did not differ between groups (P=0.160). Because sex, age, and body weight are established determinants of circulating adipokine concentrations, ²⁵ , ²⁶ we performed prespecified multiple linear regression analyses adjusting for these covariates. After adjustment for age, sex, and body weight, the observed lower leptin concentrations in the dyslipidemia group were less pronounced (adjusted P=0.057). Ferritin was not included as a covariate due to lack of evidence supporting a direct effect on circulating leptin concentrations.

Table 2.

Plasma Biomarkers on Day of Surgery

	Dyslipidemia	No dyslipidemia	P value
Lipids and lipoproteins
Total cholesterol, mmol/L	5.26 ± 1.51	5.05 ± 0.69	0.296
High‐density lipoprotein cholesterol, mmol/L,	1.02 (0.97–1.21)	1.35 (1.19–1.48)	<0.001*
Low‐density lipoprotein cholesterol, mmol/L	3.17 (2.56–4.05)	3.13 (2.67–3.59)	0.296
Triglycerides, mmol/L	1.62 (1.14–2.27)	1.24 (1.03–1.54)	0.007*
ApoAI, mg/dL	136.40 (119.68–162.57)	161.35 (146.49–177.71)	<0.001*
ApoB, mg/dL	93.15 ± 25.17	81.81 ± 16.29	0.007*
Adipokines
Leptin, ng/mL	45.18 (29.74–63.43)	66.91 (40.51–88.21)	0.010*
Adiponectin, μg/mL	5.34 (4.12–7.57)	6.08 (4.88–8.20)	0.072
Adiponectin‐to‐leptin ratio	130.71 (88.05–172.23)	97.93 (66.60–187.97)	0.160

Data expressed as mean±SD or median (first–third quartile).

Apo indicates apolipoprotein.

^*Significant findings with P<0.05.

Correlation analysis of adipokines with lipids or apolipoproteins is seen in Figures 2 and 3: Leptin showed weak positive correlations with HDL‐C (Spearman's ρ=0.24, P<0.01; Figure 2A) and its primary apolipoprotein apoAI (Spearman's ρ=0.28, P<0.01; Figure 2B). Adiponectin showed moderate positive correlation with plasma HDL‐C (Spearman's ρ=0.30, P<0.01; Figure 3A), weak positive correlation with apoAI (Spearman's ρ=0.20, P=0.03; Figure 3B) and weak negative correlations with triglycerides (Spearman's ρ=−0.19, P=0.03; Figure 3C), and ApoB (Spearman's ρ=−0.20, P=0.03; Figure 3D). No other significant correlations between adipokines and lipid or lipoprotein measures were observed. Correlation analysis of adipokines with insulin resistance (homeostatic model assessment of insulin resistance‐1) revealed a moderate inverse relationship association between adiponectin and homeostatic model assessment of insulin resistance‐1 (Spearman's ρ=−0.45, P<0.001, Figure S3), whereas no association was observed between leptin and homeostatic model assessment of insulin resistance‐1 (Spearman's ρ=0.08, P=0.63).

Figure 2. Correlation plots of leptin with HDL‐C (A) and apoAI (B) with lines of best fit and CIs.

Spearman's rank correlation (r), and its Benjamini–Hochberg adjusted P value, only significant correlations depicted. apoAI indicates apolipoprotein AI; and HDL‐C, high‐density lipoprotein cholesterol.

Figure 3. Correlation plots of adiponectin with HDL‐C (A), apoAI (B), triglycerides (C), and apoB (D) with lines of best fit and CIs.

Spearman's rank correlation (r), and its Benjamini–Hochberg adjusted P value, are the only significant correlations depicted. apoAI indicates apolipoprotein AI; apoB, apolipoprotein B; and HDL‐C, high‐density lipoprotein cholesterol.

Transcriptomics

To characterize transcriptomic pathways linked to dyslipidemia across metabolic tissues, we analyzed RNA sequencing data from SAT, VAT, jejunum, and liver tissues. We applied 2 complementary analytical strategies. First, we compared gene expression between patients with obesity and dyslipidemia and those with obesity without dyslipidemia. Second, we assessed associations between tissue‐specific gene expression and continuous plasma lipid and lipoprotein concentrations to identify transcriptomic pathways defining dyslipidemia. This approach identified a large number of genes associated with dyslipidemia status as well as genes whose expression correlated with plasma triglycerides, HDL‐C, and apoAI, whereas no significant associations were found for total cholesterol, LDL‐C, or apoB. An overview is provided in Figure 1; genes associated with dyslipidemia status are listed in Table S1 (DESeq2 analysis, FDR adjusted), and genes associated with continuous lipid or lipoprotein concentrations are listed in Table S2 (microbiome multivariable association with linear model linear modeling, FDR adjusted).

In SAT (116 patients, 40 with dyslipidemia), genes were up‐ or downregulated in association with dyslipidemia status and when comparing gene expression of SAT with continuous lipid or lipoprotein concentration, we found associations with plasma triglycerides only. Figure 1 describes under which pathway a specific gene can be categorized—AT inflammation, adipogenesis, or oxidative stress—and Tables S1 and S2 provide detailed description of individual genes. In SAT, comparison of patients with versus without dyslipidemia showed upregulation of IL32, downregulation of FOS and NFIL3, and minor downregulation of WDR74 (Figure S4A, Table S1). In sex‐stratified analyses, STAB2 and TRAF4 were upregulated in men with dyslipidemia, whereas MT1X was downregulated. Across the cohort, gene expression in SAT in relation to plasma triglycerides levels showed modest upregulation of ENDOG, NOTCH3, and UCHL1, but strong upregulation of TECRP1, and modest downregulation of mitochondrial genes, PTPRQ, and STOX1. Pathway (pathfindR KEGG) enrichment analysis of SAT genes associated with plasma triglycerides identified several overrepresented pathways related to Notch signaling, HIF‐1 (hypoxia‐inducible factor 1) signaling, oxidative phosphorylation, and chemical carcinogenesis‐reactive oxygen species (ROS) production pathways and are described in the Suporting Information (Figures S5–S7).

In VAT (124 patients, 43 with dyslipidemia), we found gene correlations with dyslipidemia status, and continuous concentration of plasma triglycerides, HDL‐C, and apoAI. Figure 1 provides an overview of pathways affected–namely, inflammation, adipogenesis, or oxidative stress–and Tables S1 and S2 provide descriptions of most genes. In patients with dyslipidemia, EBF3, TTYH1, and ZHX1 were downregulated, whereas APOA4 was upregulated (Figure S4B and Table S1). Across the cohort, several genes expressed in VAT showed associations with continuous lipid concentrations (Table S2). Positive associations with plasma triglycerides were observed for C5AR1, CCL3, CCR7, FGR, MSR1, NLRP3, and UCHL1. AZGP1, CALCRL, and SLC2A4 showed positive correlation with HDL‐C, whereas EGFL6 had a negative correlation with HDL‐C. TXNRD1 showed a minor negative association with HDL‐C. VAT, pathfindR KEGG analysis in relation to plasma triglycerides revealed enrichment of several genetic pathways and is described in the Supporting Information.

In liver tissue (123 patients, 41 with dyslipidemia), transcriptomic analysis and pathfindR KEGG analysis found not only genes involved in de novo lipogenesis and lipid metabolism (steroid biosynthesis; 60.0‐fold enrichment, unsaturated fatty acid biosynthesis; 46.7‐fold, PPAR [peroxisome proliferator‐activated receptor] signaling; 15.2‐fold) but also genes associated with liver steatosis or inflammation as depicted in Figure 1. In the full cohort, positive associations with plasma triglycerides were observed for FASN, FDFT1, ME1, and SQLE. GDF15 showed a positive association with plasma apoAI, whereas SERPINE1 and STAT4 were negatively associated with plasma apoAI. Genes are further described in Tables S1 and S2.

In the jejunum (118 patients, 41 with dyslipidemia), only 2 genes were differentially expressed between patients with and without dyslipidemia: HOXC6 and LCN2, both of which were upregulated in dyslipidemia (Table S1). No associations were found with continuous lipid or apolipoprotein concentrations.

Metabolomics

Several plasma metabolites were associated with dyslipidemia, as shown in Figure 4. Palmitoyl ethanolamide and oleoyl ethanolamide, both endocannabinoid‐like lipid metabolites and PPARα agonists, were positively associated with dyslipidemia. The GM‐derived secondary bile acid isoursodeoxycholate (isoUDCA) was also positively associated with dyslipidemia, whereas sphingolipids and plasmalogens showed negative associations. In total, 234 metabolites were associated with continuous concentrations of plasma lipids, lipoproteins, or adipokines (Figures S8 and S9). Most associations with plasma lipids were positive, including those involving monoacylglycerols and lipid derivatives. Some lipid metabolites, however, showed negative associations, including butyrate‐related metabolites, sphingolipids, and plasmalogens, all of which were inversely associated with plasma triglycerides concentrations. Several metabolites were also associated with adipokines, including butyrates, which showed negative correlations with leptin.

Figure 4. Metabolites in dyslipidemia.

The coefficient refers to the MaAsLin2 correlation coefficient to dyslipidemia status; *P<0.05; **P<0.01; ***P<0.001. MaAsLin2 indicates microbiome multivariable association with linear models; SAT, subcutaneous adipose tissue; and VAT, viscous adipose tissue.

Multiomics Integration of Adipose Tissue Transcriptomics and Plasma Metabolomics

Correlation‐based network analysis incorporating SAT and VAT transcriptomics and plasma metabolomics found a limited number of significant cross‐omics associations (Figure 5), including inverse relationships between SAT NFIL3 and isoursodeoxycholic acid and between VAT EBF3 and N6‐methyllysine, as well as positive associations between SAT WDR74 and VAT ZHX1 and COL4A3BP.

Figure 5. Multiomics correlation network.

Network of significant correlations between adipose tissue gene expression and plasma metabolites in dyslipidemia. Node color denotes tissue or metabolite type. Edge color and intensity reflect the direction and magnitude of correlations. PPARα indicates peroxisome proliferator‐activated receptor alpha,

Gut Microbiota Analyses

Gut metagenomic analyses were exploratory and focused on taxonomic relative abundance. Only a modest difference in GM composition (relative abundance) between patients with and without dyslipidemia was found, results of which are detailed in Supporting Information Results and Figure S10–S12. On a bacterial order level, in patients with dyslipidemia, we found more species of the order Lachnospirales and of the order Peptostreptococcales, both part of the Firmicutes phylum. In patients without dyslipidemia, Oscillospirales (also within the Firmicutes phylum), Bacteroidales, and Lactobacillales predominated (Figure S12).

DISCUSSION

We investigated 125 individuals with obesity of whom about one‐third had dyslipidemia. Individuals with obesity and dyslipidemia had higher triglycerides and apoB and lower HDL‐C and apoAI, a typical pattern of dyslipidemia seen in obesity. ²⁷ In unadjusted analyses, leptin concentrations were lower in dyslipidemia, whereas adiponectin and adiponectin‐leptin ratio did not differ. However, after adjustment for age, sex, and body weight, the difference in leptin concentrations between groups was reduced, suggesting that plasma adipokines may have limited independent association with dyslipidemia in obesity. The observed inverse association between adiponectin and insulin resistance is consistent with established literature ²⁸ and likely reflects known metabolic physiology rather than a novel finding.

Next, in contrast to our hypothesis of the predominant role of VAT in dyslipidemia, our transcriptomics revealed that both VAT and SAT had altered transcriptome in relation to dyslipidemia or plasma lipids, and notably, SAT had most numerous gene alterations. We found that genes were up‐ or downregulated in relation to plasma triglycerides, HDL‐C, and apoAI rather than cholesterol or apoB. Because previously mainly VAT associated with cardiometabolic disease, ¹⁰ , ¹¹ , ¹² our finding of altered SAT transcriptome in dyslipidemia underscores an underestimated role of SAT in dyslipidemia pathogenesis. We identified 3 recurring pathways in AT in relation to dyslipidemia or lipid profiles: (1) AT inflammation, (2) adipogenesis, and (3) oxidative stress.

AT inflammation has been described as a factor in cardiometabolic disease development. ⁶ , ¹³ In our study, several inflammatory pathways in AT were associated with dyslipidemia. In SAT, the proinflammatory cytokine gene IL32 was upregulated, whereas NFIL3 was downregulated in patients with dyslipidemia. NFIL3 encodes E4bp4, a transcription factor maintaining anti‐inflammatory AT homeostasis through IL‐10 (interleukin‐10)‐mediated M2 macrophage polarization and promotion of regulatory T cells. ²⁹ , ³⁰ Downregulation of NFIL3 in dyslipidemia may therefore reflect a shift toward a more proinflammatory immune milieu in SAT. We found strong STAB2 upregulation in men with dyslipidemia; STAB2 is a scavenger receptor indirectly regulating inflammation through affecting clearance of proinflammatory ligands. ³¹ Although we found only minor downregulation of the TGF‐β (transforming growth factor beta) coactivator gene, WDR74, in dyslipidemia, TGF‐β is essential for adipogenesis and immune homeostasis, and impaired TGF‐β has been shown to contribute to inflammatory pathologies. ³² In both SAT and VAT, UCHL1 expression correlated to elevated triglycerides, in line with previous reports that UCHL1 is upregulated in AT of patients with obesity and increases inflammation through degradation of RXRα (retinoid X receptor alpha) and inhibition of LXR (liver X receptor)/RXR's anti‐inflammatory effects. ³³ In VAT, proinflammatory CCL3 positively correlated with plasma triglycerides, and CALCRL had minor positive association with HDL‐C. CALCRL encodes adrenomedullin receptor with anti‐inflammatory effects through NF‐κB (nuclear factor kappa B) inhibition, and reduces inflammatory cytokines and CRP (C‐reactive protein), whereas CCL3 (C‐C motif chemokine ligand 3) is released by inflamed AT in obesity, working as a chemoattractant for monocytes and macrophages partly through NF‐κB. ³⁴ , ³⁵ NF‐κB signaling could be a driver of AT inflammation and ROS production during overnutrition and metabolic dysregulation. ³⁶ In VAT, AZGP1, had minor positive correlation with HDL‐C and adiponectin, AZGP1 (zinc‐α‐2‐glycoprotein) is an adipokine increasing lipolysis, AT browning, and insulin sensitivity by, among others, suppression of proinflammatory pathways involving NF‐κB, IL1β, and IL‐6. ³⁷ In VAT, CCR7 had positive association with plasma triglycerides, which could be in line with its previously described role in coordinating immune cell trafficking and sustaining AT inflammation in obesity. ³⁸ Additional inflammatory genes in VAT, including C5AR1 and NLRP3 showed statistically significant but modest positive associations with plasma triglycerides and should be interpreted with caution. C5AR1 is involved in complement‐mediated M1 macrophage recruitment, ³⁹ whereas NLRP3 regulates inflammasome activation and IL‐1β and IL‐18 production in AT. ⁴⁰

Adipogenesis emerged as the next key transcriptomic pathway. In SAT, FOS, a critical regulator of preadipocyte differentiation through its interaction with AP‐1 (activator protein 1), ⁴¹ showed pronounced downregulation in dyslipidemia. PTPRQ, another regulator of adipogenesis via the PI3K‐PKB/Akt pathway, ⁴² correlated negatively with triglycerides. In men with dyslipidemia, TRAF4 was upregulated in SAT, consistent with its function as adipogenesis checkpoint regulator. ⁴³ In VAT, EGFL6 expression negatively correlated with HDL‐C; a study analyzing AT in children with obesity identified EGFL6 as a key secreted extracellular matrix protein with upregulation during adipocyte differentiation and remodeling, and strong associations with adipose hypertrophy, macrophage infiltration, high sensitivity‐CRP, and insulin resistance. ⁴⁴ Its negative correlation with HDL‐C, a protective lipoprotein, may reflect altered adipocyte development in dyslipidemia. Other genes with roles in adipogenesis, including NOTCH3, EBF3, and SLC2A4, showed statistically significant but comparatively modest associations. Overall, these findings indicate that dyslipidemia is associated with up‐ or downregulation of adipogenic pathways, consistent with complex alterations in AT remodeling.

Third, consistent with earlier findings, ⁴⁵ dyslipidemia was associated with transcriptomic signatures related to oxidative stress (ROS production), mitochondrial function, and hypoxia signaling (HIF‐1) in AT. Autophagy is essential for the cellular response to oxidative stress, ⁴⁶ , ⁴⁷ and in SAT we observed strong positive association between TECRP1 expression and plasma triglycerides. This is relevant because TECRP1 produces one of the main autophagy tethering proteins facilitating autophagosome‐lysosome fusion to regulate ROS. MT1X, a metallothionein protecting against oxidative stress through free radical scavenging, ⁴⁸ was markedly downregulated in SAT of men with dyslipidemia, which might have indicated reduced oxidative stress resilience in AT dysfunction. ENDOG, the gene for a mitochondrial nuclease involved in autophagy regulation and oxidative stress response, ⁴⁹ also positively correlated with plasma triglycerides in SAT. We identified genes of oxidative phosphorylation, ROS production, and HIF‐1. UCHL1 expression in SAT and VAT had positive relation with triglycerides, UCHL1 modulates oxidative stress response through LXR/RXR signaling. ³³ STOX1, a transcription factor implicated in oxidative stress regulation also expressed in AT, ⁵⁰ , ⁵¹ showed a negative association with plasma triglycerides in SAT. Lastly, in VAT, FGR showed a positive, lower‐magnitude association with plasma triglycerides. FGR is a downstream effector of ROS signaling and has been implicated in proinflammatory M1‐like macrophage polarization in AT, driven by mitochondrial ROS and complex II activation. ⁵² Next, our PathfindR KEGG analyses suggested overrepresentation of hypoxia‐responsive (HIF‐1 pathways) and mitochondrial oxidative phosphorylation pathways in SAT in relation to plasma triglycerides. However, given the modest effect sizes and bulk‐tissue transcriptomic design, these findings should be interpreted as descriptive patterns rather than evidence of specific functional alteration.

Hepatic transcriptome in relation to dyslipidemia or plasma lipids confirmed alterations of genes involved in lipid metabolism and de novo lipogenesis in relation to plasma triglycerides but also highlighted genes with roles in hepatic steatosis or inflammation (metabolic‐associated fatty liver disease or metabolic dysfunction‐associated steatohepatitis) including: FASN, GDF15, ME1, SERPINE1, SQLE, and STAT4. SQLE, the gene for squalene epoxidase is the second rate‐limiting enzyme of cholesterol synthesis and has been found to contribute to hepatic steatosis and metabolic dysfunction‐associated steatohepatitis. ⁵³ SLC2A8, encoding a fructose‐glucose transporter GLUT8 also described in liver steatosis. ⁵⁴ The lack of correlation between LDL‐C and hepatically expressed genes (including HMGCR, LDLR, LDLRAP1), may reflect that, in obesity, dyslipidemia is more commonly characterized by elevated triglycerides and reduced HDL‐C, with more modest changes in LDL‐C. ⁵⁵

Our metabolomics in dyslipidemia found a correlation with palmitoyl ethanolamide and oleoyl ethanolamide, both endocannabinoid‐like lipid metabolites and PPARα agonists implicated not only in energy‐ and lipid metabolism but also in inflammation and cardiovascular disease. ⁵⁶ , ⁵⁷ Oleoyl ethanolamide is likely involved in lipid metabolism through lipolysis enhancement ⁵⁸ but is also described providing crosstalk between intestinal microbiota and intestinal homeostasis, where its malfunction is associated with enteropathies, obesity, and chronic inflammation. ⁵⁹ Studies have shown that palmitoyl ethanolamide has anti‐inflammatory action and can attenuate atherosclerotic plaque formation, decrease fat accumulation and lower plasma lipids. ⁶⁰ , ⁶¹ , ⁶² Next, we found positive correlation between isoUDCA and dyslipidemia; isoUDCA is a GM‐derived bile acid, and a recent study already postulated its influence on lipid metabolism. ⁶³ Sphingolipid metabolism also appears to influence lipid metabolism and cardiometabolic disease, ⁶⁴ we found negative relation between sphingolipid metabolites and dyslipidemia. Butyrate‐related metabolites, besides being endogenously produced from normal lipid metabolism, are short‐chain fatty acids (SCFAs) produced by the GM. SCFAs are increasingly investigated for their protective or anti‐inflammatory roles and ability to reduce dyslipidemia, atherosclerosis and CVD risk ⁶⁵ , ⁶⁶ ; however, in our analysis these metabolites showed negative correlations with HDL‐C and apoAI; thus the beneficial aspect is less clear here but may indicate a regulatory function through HDL‐C. Plasmalogens negatively associated with plasma triglycerides, and a previous study has identified reduced plasmalogens in cardiometabolic disease. ⁶⁷ Plasmalogens appear to regulate thermogenic fat activation by induction of mitochondrial fission, ⁶⁸ potentially increasing fatty acid oxidation and reduce plasma triglycerides.

Multiomics integration of AT transcriptomics with plasma metabolomics identified a limited number of cross‐layer associations, most notably an inverse association between SAT expression of NFIL3, a regulator of AT immune homeostasis and circulating levels of the GM‐derived secondary bile acid isoUDCA, which positively associated with dyslipidemia.

Gut metagenomics revealed only modest differences in relative bacterial abundance between patients with and without dyslipidemia, with no significant difference in alpha or beta diversity. Differences were limited to the bacterial order level and were of small magnitude. Although previous research has linked SCFA‐producing bacteria primarily to metabolic health, ⁶⁵ , ⁶⁶ in our results SCFA‐producing bacterial orders were observed in both patients with dyslipidemia (Lachnospirales) and those without dyslipidemia (Oscillospirales, including butyrate‐producers). These findings indicate that different bacterial taxa, whether SCFA producing or not, are present across dyslipidemia status and do not define a uniform microbial signature of dyslipidemia.

Our study carries limitations. First, analyses were restricted to individuals with obesity undergoing bariatric surgery with a BMI of at least ≥35 kg/m², and with a predominance of female participants, which may limit generalizability across the full BMI spectrum. Second, the cohort was predominantly White, potentially limiting applicability to more ethnically diverse populations. Third, the cross‐sectional design and the fasting perioperative sampling context may have affected metabolomic and transcriptomic profiles and limit causal interpretation. Finally, although BMI was comparable between groups, residual effects of body weight itself cannot be fully excluded. Accordingly, the observed molecular differences should be interpreted as associations with dyslipidemia rather than evidence of mechanistic effects.

Conclusions

In conclusion, our study provides a well‐characterized cohort with multiomics profiling offering novel insights into dyslipidemia in obesity. In line with previous studies, ¹⁰ , ¹¹ , ¹² our findings support a potential role of VAT in dyslipidemia, while also highlighting an underappreciated role of SAT in lipid metabolism and dyslipidemia. Across ATs, dyslipidemia was consistently accompanied by differential transcriptomic expression patterns related to inflammation, oxidative stress, and adipogenesis, suggesting that these pathways may collectively characterize a spectrum of AT health versus dysfunction in obesity. Plasma metabolomics further identified coordinated alterations in lipid‐related metabolites, including endocannabinoid‐like metabolites and GM‐derived isoUDCA and butyrate‐related metabolites, and our targeted transcriptomics‐metabolomics integration identified an inverse association between NFIL3 expression in SAT and circulating isoUDCA. Although causal relationships cannot be inferred, our findings offer an exploratory transcriptomic and metabolomic overview of molecular patterns associated with dyslipidemia in obesity. Future studies across broader BMI ranges and functional validation will be needed to clarify the relevance of these associations and to determine whether the identified pathways and metabolites have prognostic or therapeutic value for treating dyslipidemia and cardiovascular disease.

Sources of Funding

Max Nieuwdorp was supported by a Novo Nordisk Foundation GUTMMM consortium grant 2016 [NNF15OC0016798], on which Max S. Z. Zwartjes was appointed. Max Nieuwdorp is supported by a personal NOW VICI grant 2020 (09150182010020) and an European Research Council Advanced grant (101141346). Abraham S. Meijnikman is supported by a personal ZONMW‐VENI grant 2023 (09150162310148). The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Disclosures

Max Nieuwdorp is founder and a scientific advisory board member of Caelus Pharmaceuticals and Advanced Microbiome Interventions, the Netherlands. He is also on the board of directors of Diabeter Netherlands BV. However, these conflicts of interest are not directly related to the content of the current article. The other authors declare no competing interests.

Supporting information

Tables S1–S2

Figures S1–S12

Data S1

References 72‐82

STROBE Statement